315 – Sampling Strategies for Rare and Hard-to-Reach Populations
Revisit Sample Size Estimation in Phase II Selection Designs
Zuoshun Zhang
Celgene Corporation
Angela Hu
Celgene Corporation
The statistical methods for ranking and selection were introduced and used in the design of phase 2 oncology clinical trials, where subjects were randomized to several promising treatment arms with the goal to select one arm for further development. The methods were generalized to selection designs with time-to-event endpoint and different designs with binary endpoint. In order to facilitate its wider application, there need readily applicable methods for sample size calculation. For classical selection designs with binomial endpoint, we show that the sample size can be calculated exactly using a SAS program. For selection designs with time-to-event endpoint, we adapt Bechhofer's method for normal endpoint and estimate sample size of total number of events; and we further show via simulation the designs have approximately the specified correct selection probability. For a class of flexible selection designs with binary endpoint, we point out a possible flaw of the minimum advantage requirement based on response rate, and propose a new class of designs based on the number of observed responses and calculate sample size with specified correct selection probability.